Linux-Development-Sys Digest #782, Volume #6 Fri, 4 Jun 99 22:14:22 EDT
Contents:
Re: can 2 linux machines talk to a dual port scsi raid system ?? (Andreas Dilger)
Re: What are the differences between mySQL and mSQL? (Don Baccus)
Re: TAO: the ultimate OS (Christopher Browne)
Re: What are the differences between mySQL and mSQL? (Don Baccus)
Large versions of Linux (bill davidsen)
Re: Q: Linux equivalent technologies to Win32? (Selious)
----------------------------------------------------------------------------
From: [EMAIL PROTECTED] (Andreas Dilger)
Subject: Re: can 2 linux machines talk to a dual port scsi raid system ??
Date: 5 Jun 1999 00:41:36 GMT
In article <[EMAIL PROTECTED]>,
dan <[EMAIL PROTECTED]> wrote:
>i have a dual port scsi raid system with 2 linux systems connected to it
>
>both run redhat 5.2
>there is an ext2 filesystem on the raid box (~26Gbytes)
>the problem i have is that if linux box A writes a file to the raid
>filesystem,
>linux box B doesn't see the file
I assume, first of all, that you are ONLY READING from system B. Otherwise,
you are in for a world of hurt as system B overwrites data and inodes written
by system A, since there is no synchronization between the two. In fact, on
other clustered systems I have seen (AIX, HPUX), only one system has an
active connection to the shared disk, and only at failover does the backup
system even mount the filesystems... Most filesystems aren't designed to
even have two systems reading from the same disk, as they do lazy writes,
and don't re-read data that they have cached.
I believe that you need to do something on system B to flush the "file" cache
before mounting the shared disk. This is because Linux will buffer access to
a block device, even if it is accessed direcly rather than through the
filesystem layer. This was a complaint of database vendors about Linux - no
RAW filesystems. I'm not sure of a 100% reliable way of flushing the cache,
but you can read a bunch of files from disk, or have a small program to malloc
(and modify) an amount of memory comparable to the total amount of RAM on
system B. Maybe there's a kernel API, but I'm no kernel guru.
Cheers, Andreas
--
Andreas Dilger University of Calgary \"If a man ate a pound of pasta and
Micronet Research Group \ a pound of antipasto, would they
Dept of Electrical & Computer Engineering \ cancel out, leaving him still
http://www-mddsp.enel.ucalgary.ca/People/adilger/ hungry?" -- Dogbert
------------------------------
Crossposted-To:
comp.os.linux.development.apps,comp.os.linux.misc,comp.os.linux.setup,comp.lang.java.databases
Subject: Re: What are the differences between mySQL and mSQL?
From: [EMAIL PROTECTED] (Don Baccus)
Date: 4 Jun 1999 17:34:23 PST
In article <[EMAIL PROTECTED]>,
Kaz Kylheku <[EMAIL PROTECTED]> wrote:
>On 3 Jun 1999 09:30:32 PST, Don Baccus <[EMAIL PROTECTED]> wrote:
>No, in this culture free means ``comes source code that you can redistribute
>and modify, or both''.
Sybase for Linux is (currently) free. Live with it.
It's not Open Source and no one is forcing you to use it
(personally, I use Postgres for my personal needs precisely
because it is open source) but it is free. As defined in
english.
--
- Don Baccus, Portland OR <[EMAIL PROTECTED]>
Nature photos, on-line guides, at http://donb.photo.net
------------------------------
From: [EMAIL PROTECTED] (Christopher Browne)
Crossposted-To: comp.os.linux.advocacy,comp.os.misc
Subject: Re: TAO: the ultimate OS
Reply-To: [EMAIL PROTECTED]
Date: Sat, 05 Jun 1999 00:30:54 GMT
On Fri, 04 Jun 1999 15:34:24 -0400, G. Sumner Hayes
<[EMAIL PROTECTED]> wrote:
>"Vladimir Z. Nuri" wrote:
>> I'm not designing memes? excuse me? anyone who posts on
>> usenet is designing memes, imho.
>
>Curious opinion. I can see an argument that anyone who posts
>is spreading memes, but "designing" implies originality and
>not just repeating things that have been long studied and either
>are well-known or long since rejected.
It can certainly be original to use "old" ideas when describing a new
system, but in order for this to be of value, it is necessary for the
new material to represent a fairly 'deep' "re-presentation" of the
juxtaposition of the old ideas.
Commentary to the effect that "It's got to be object oriented" and
"It's got to be user-friendly" do not provide a sufficiently deep
presentation to be satisfactory.
>> where are the documents?
>
>I gave you a number of pointers, both to names and URLs of
>institutions with literally hundreds of documents.
>
>> HCI.. what is that?
>
>Human-computer interaction. It's the field dedicated to making
>it easier for people to get computers to work for them. It includes
>psychological research to determine how people think about solving
>a task, "normal" UI development, hardware development, and anything
>else that can make computers easier to use.
>
>Given that you say:
>
>"The system favors end-user convenience and simplicity more than
>any other criteria"
>
>HCI should be the primary focus of your research. But you show
>a complete lack of understanding of the field. I really think
>you should educate yourself in the current state of the field before
>you try to duplicate the work of hundreds of people on your own.
>Courses or self-study, I don't care -- it'll take a while to get
>up to speed, but if you come in understanding usability issues
>you'll have at least a chance of designing a system that meets the
>goal above.
Indeed.
<http://www.sylvantech.com/~talin/projects/ui_design.html>
and
<http://www.mit.edu/~jtidwell/interaction_patterns.html>
represent some fairly decent brief surveys on issues of user interface
design.
It is not at all obvious that these sorts of issues need to couple
deeply with device drivers, memory management, process management, and
file system management, that are the "bread and butter" of the
resource allocation for which an OS is responsible.
>> : judging from the reaction you're getting you need to change your
>> : approach if you really want to convince people.
>>
>> what approach do you suggest, dear sir?
>
>Well, I can think of a number:
>1. Show some code -- that seems to be what people here want.
>2. Post an original, well-researched, and concise design
> document.
>3. Move to a different forum that might be more receptive.
>
>That is assuming that you're actually interested in coming up
>with a better system. If you just like flamewars then your
>approach is doing fine.
It would be perhaps more appropriate to go to comp.windows.x, or
comp.windows.x.kde, or something of the sort, if the issue is of how
to grapple with HCI issues on a UNIX-like OS.
If the issue is of how to start design and implementation of a new
"Truely Object Oriented OS" (spelling daemons beware; any misspellings
here are purely intentional...), then comp.os.research or comp.os.misc
are appropriate fora.
It is largely a waste of time to try to change the direction of the
low level design of either Linux or UNIX, as neither have a mandate at
the level of those that make *sweeping* design decisions that will
allow such efforts to be other than wasted breath.
>> do you believe it is a worthwhile goal to come up with a design
>> document that represents a consensus, irrespective of code, prior
>> to code being written?
>
>Yes (unless there's already such a document out there). But it needs
>to be a document that is well-studied and probably has tested some of
>the ideas with sample implementations.
It doesn't make sense for (for instance) the Linux community to make
the sorts of sweeping changes that are implied without there being
some usable prototype code available to establish that the design
approach has at least enough merit that it can cope with the creation
of a prototype.
The GnuCash project has been grappling with problems not dissimilar to
this; its codebase has grown somewhat unwieldy as it has accreted
fragmentary portions of support for:
a) Motif,
b) GNOME,
c) Qt,
d) An HTML library for reporting,
e) Embedded Scheme for scripting,
f) Embedded Perl for scripted reports.
That's not *all* of the components; there are more than that.
Essentially, a bunch of people came in, and had some particular ideas
for their "pet areas" of the project, and things just added to it to
the point now where the main developers are trying to do some
"rationalization" (in ManagementSpeak) to take out some of this stuff
so as to make it possible for people to actually compile and run it.
(Guile will probably remain; ePerl will probably get nuked; the
running battle at this point is GNOME vs Motif...)
The point of bringing this up is that the GnuCash project has suffered
from a bit of lack of coherent design, with the result that the
software hasn't been able to stabilize to the point of permitting a
non-alpha release. They were, for too long, happy enough to add in
whatever anyone wanted; there wasn't a mandate to say "No." There
seems now to be some consensus that components that aren't
contributing to the stability of the design may need to get dropped,
at least for now.
In contrast, the Linux kernel folk and UNIX folk are not likely to be
nearly so friendly about the matter; a loud "No" *will* resound when
attempt is made to make sweeping changes that would make stability of
existing systems suffer.
Linus said "no" when the GGI folk wanted to toss a bunch of graphical
stuff in the kernel. This caused some consternation and ill will in
both directions; retrospect seems to be showing that that decision was
the correct one at the time.
Some framebuffer support has recently been added to the kernel, which
should provide the kernel functionality that GGI needs.
The fact that they had to cope with *not* having the ability to
intimately fiddle with kernel functionality forced them to make the
libraries *greatly* more portable, which will doubtless make it
*vastly* easier to deploy GGI not only on Linux but also on other
OSes, and *even atop X11.*
>> : For example, do you understand why there was a boom in object
>> : databases (like you propose for the filesystem) about 5 years ago
>> : and why people have become much less interested in the idea today?
>>
>> performance?
>
>No.
>
>> : Or the FS approach that Beos took? Do you understand the difference
>> : between the object models in C++, Python, and Smalltalk? How about
>> : signatures? Information-flow (leaving that term nebulous) interfaces
>> : as an alternative to object-oriented (idem) design?
>>
>> there are many ways to implement objects. I'm familiar with the
>> basics (including a more recent language you don't mention, java)
>> notice that you are referring to languages, not OSes.
>
>No, I'm referring to object models. That's why I said object models.
>Duh. I don't mention Java because it uses the same basic object model
>as one of the languages I do mention. I also don't mention CLOS,
>Perl5, Objective C, Ada95, Modula, and a bunch of other languages that
>use basically the same object models as the languages above or other
>less-common object models.
>
>In your document you say things that show that you really don't
>understand common object models -- for instance, you conflate
>"C++-style object" with "interface", which shows that you either
>don't understand the C++ object paradigm or you don't know what
>interface means in the OO community.
Unfortunately, it is all too common for people that think they
understand OO to, in reality, not understand it. This is obviously
true in cases where they are not actually familiar with any OO
implementation, as is the case when managers bleat: "We've got to make
it Object Oriented!" so as to conform with the other managers in the
industry.
It is also true when people are only familiar with one approach to OO,
as is too often the case for users of C++, who seem often enough to
confuse the use of C++ with "object orientedness," when C++ is
*actually* a language that supports *various* paradigms of which OO is
only one.
>> : Just screaming "make it all object oriented!" is worthless. It's
>> : not even clear what you mean by object oriented, but if you mean
>> : bundling data and code tightly together then there's a pretty
>> : strong consensus among HCI researchers that that is a terrible way
>> : to design a system if you're worried about the end user being able
>> : to do what he's trying to do without a fuss.
>>
>> the point is to make the entire system based on objects
>
>But you haven't clearly defined what an object is. You've
>listed a few properties of objects, but not enough to give the
>impression that you've seriously considered the options and
>made informed choices.
>
>Let's dissect the parts you give to an object:
>
>: An object can be contained in other objects
>
>IOW, objects are first-class elements of the system. There are no
>exceptions. I'm not sure how you plan to do this efficiently, but
>at least it's well-formed.
>
>: An object has several basic elements: a *name*, *type*, *version*
>: *interface*, *code*, and *data*
>
>By "name" do you mean a locater in an organized namespace used to
>open communications to the object (as per Plan 9 names) or do you
>mean a reference that's a unique handle to an object (as per Java
>references)?
>
>By "type", what do you mean? Are types a string explaning what
>kind of thing the object is as per MIME types? Are they a hint to
>the system as to what to do with the object as per C++ types? Are
>they an absolute guarantee as per ML types? Are types dynamic or
>static? Strong or weak?
>
>By "version" what do you mean? Is this a guarantee of a specific
>interface a la HTTP versions? A compatibility measure as per
>glibc symbol versions or kernel module versions? A hint of some
>sort about stability of the interface?
>
>When you say "interface" do you mean just an API/ABI? Or do you
>mean something like OO interfaces, an inherited set of calls that
>defines an object hierarchly? If the latter, why is only one
>interface allowed? If multiple interfaces are allowed, you should
>clarify that. If inheritance is possible, you should specify how.
>
>By "data" do you mean arbitrary bits? A typed byte stream? A list
>of objects? Why is all data associated one-to-one with code? What
>mechanisms exist for having multiple views (possibly from disparate
>machines with disparate capabilities, possibly in different modes)
>onto the same data given that data has a bijective relationship with
>code? If there is some such mechanism, what is the special nature
>of the code that's in the same object as the data?
Note that the word "properties" would be a vastly better way to
characterize these properties, rather than describing them as
"elements."
>By "code", do you mean actual machine instructions? Source-level
>code to be interpreted? If the former, how do you propose
>that:
>
>: Objects can be "moved around" to different "places" either on a
>: local computer or remote computers.
>
>Is that only within heterogenous computers? Or are all objects
>guaranteed to have binary representations available on some set
>of architectures? Or is source code included so that the objects
>can be built natively as they migrate? Or is there a standard
>architecture a la the JVM that is abstracted out on all
>architectures?
>
>All of these issues have huge amounts of literature behind them, and
>yet you make no effort to discuss them. Throwing a word like "type"
>or "interface" out there and then not discussing the basics of it
>is unprofessional at the least.
>
>> I don't know what you mean by "bundling code and data together"
>
>That's a part many object models. Indeed, you even list code and
>data as two of the parts of your objects.
Ever since the beginning, the "big deal" about OO has always been that
of providing an association between code and data.
It might be a more expressive use of English to characterize this
using the phrase "providing a taxonomy that associates data elements
with the methods used to manipulate them."
If that isn't understood, then I'd have to question whether OO is
truly understood, as that is the basis of an Object Orientation.
>> : There are conventions, CS departments, and an ACM SIG dedicated
>> : to the general topic that you're trying to tackle. Attacking
>> : the problem without understanding even the basic research of the
>> : last 20 years is foolish. I'm not about to give an HCI course
>> : on Usenet, but CHI and Carnegie Mellon's HCI institute are good
>> : starting places that link to other sites and a good number of papers.
>>
>> believe we are talking about two different problems. I am not
>> trying to create a better object oriented paradigm for languages.
>
>Nor am I, and I'm not suggesting that you should.
>
>> I am trying to leverage the already outstanding research into
>> objects to invade the OS itself,
>
>How can you leverage that research if you haven't read it and don't
>understand it? That's my whole point -- leveraging that research is
>exactly what I would love to see, but you've demonstrated a complete
>lack of understanding of the past 20 years of research in both
>programming models (e.g. object orientation, functional programming,
>structural programming, information-flow programming) and HCI (and
>the two are sometimes closely entwined).
Indeed.
--
"Besides a mathematical inclination, an exceptionally good mastery of
one's native tongue is the most vital asset of a competent programmer."
-- Edsger W.Dijkstra
[EMAIL PROTECTED] <http://www.ntlug.org/~cbbrowne/lsf.html>
------------------------------
Crossposted-To: comp.os.linux.development.apps,comp.os.linux.misc
Subject: Re: What are the differences between mySQL and mSQL?
From: [EMAIL PROTECTED] (Don Baccus)
Date: 4 Jun 1999 17:41:13 PST
In article <3d_53.15457$[EMAIL PROTECTED]>,
Christopher Browne <[EMAIL PROTECTED]> wrote:
>The "free of cost" sense that is true for some ways that one might use
>Sybase's Linux release only expresses one meaning out of those 20-odd.
Ahhh...apparently you didn't find my sentence ambiguous.
>Are you sure the meaning is as simple as you are claiming? Or should
>you perhaps consult the OED?
Why don't you post a definition of "free" that you think applies
to software distributed under the GPL and we'll see if it really
fits.
(the sense of "unfettered", as one poster suggested, doesn't
apply - GPL-distributed software is hardly unfettered).
Now, as it happens I very much like the GPL. Nor do I use
Sybase, I use open-source Postgres for my personal work.
But, when I posted that Sybase is free on Linux, I do believe
that the meaning was clear.
--
- Don Baccus, Portland OR <[EMAIL PROTECTED]>
Nature photos, on-line guides, at http://donb.photo.net
------------------------------
From: [EMAIL PROTECTED] (bill davidsen)
Subject: Large versions of Linux
Date: 4 Jun 1999 16:32:01 GMT
I haven't visited this in a while, the last time I looked patches were
needed for (a) physical memory>1GB, (b) more than 1024 file descriptors,
and (c) more than 512 processes.
So my question is, does someone have a pointer for these patches for
2.2.9, and could this be made a kernel build option in 2.3.
Yes, I know there is a patch for up to 3.8GB floating around, but this
would be a good place to have a compile parameter. I looked at the code,
short of a major hack I don't see it becoming a tunable parameter in
/proc, as nice as that would be! Of course add space up to N on demand
would be nice, but I don't think there's a real demand for it. I would
like to be able to build a really tiny kernel, though.
--
bill davidsen <[EMAIL PROTECTED]> CTO, TMR Associates, Inc
The Internet is not the fountain of youth, but some days it feels like
the fountain of immaturity.
------------------------------
From: Selious <[EMAIL PROTECTED]>
Subject: Re: Q: Linux equivalent technologies to Win32?
Date: Sat, 05 Jun 1999 03:44:26 +0200
>
> 2. DCOM (is it RPC, RMI in Java?)
No this is the corba here !!
------------------------------
** FOR YOUR REFERENCE **
The service address, to which questions about the list itself and requests
to be added to or deleted from it should be directed, is:
Internet: [EMAIL PROTECTED]
You can send mail to the entire list (and comp.os.linux.development.system) via:
Internet: [EMAIL PROTECTED]
Linux may be obtained via one of these FTP sites:
ftp.funet.fi pub/Linux
tsx-11.mit.edu pub/linux
sunsite.unc.edu pub/Linux
End of Linux-Development-System Digest
******************************
